Why projecting regional precipitation trends is difficultChris Bretherton
Department of Atmospheric SciencesUniversity of Washington
A1B 2080-2099 minus 1980-1999
Stippling:80% of CMIP3 models agree on sign of trend
IPCC 2007
The gist of this talk
Focus: Zonally asymmetric regional precipitation trends driven by well-mixed greenhouse gas increases.
Won’t discuss, but important (Yi Ming’s talk): Precipitation is sensitive to uncertainties in regional aerosol emissions and their pattern of direct + indirect (cloud-related) radiative forcing
Observationalcontext – precip trends
visible, but lots of natural variability
IPCC 2007
Main points of this talk
1. Tropical precipitation, circulation and SST depend on poorly modeled processes such as cumulus convection.
2. Teleconnection of tropical precipitation changes to midlatitudes also varies between models.
3. Land surface and clouds feed back on precipitation changes
4. Precipitation trends modulated by overall global warming, so dependent on emissions changes and climate sensitivity.
Bottom line: Strong internal feedbacks in the climate system make precipitation trends very sensitive to model details.
1. Tropical precipitation challenges
Trends in tropical drying on ITCZ margins… quite diverse between models
Neelin et al. 2006
Precipitation trends are not just noise...each model has its own systematic response
Neelin et al. 2006
Evaporation increases over all low-lat oceans
Climate model rainfall sensitive to SST couplingCAM3
CAM3 + ocean
Coupled dSST
Double-ITCZ bias in rainfall and SST in coupled model.
Plots: NCAR CESM web site
Precip trends don’t require SST gradient trends
• Climate models simulate different tropical rainfall responses even to a uniform +2K SST increase, including over land.
Bony et al. 2004
-w500 is a rainfall proxy (10 units ~ 1 mm/d)
Tropical rainfall pattern sensitive to Cu param.
CAM3 simulations with different cumulus parameterizations have different rainfall biases
2. Teleconnection challenges
Midlatitude circulation responds more strongly to tropical SST anomalies than midlatitude SST anomalies
27% of winter ensemble-mean PNA variability explained by SSTA; of that 70% is explained by tropical SSTAs.
PNA
Lau and Nath 1994
GOGA TOGA
Tropical teleconnections depend on the subtropical jet structure, which is model-dependent
200 hPa circulation response of 3 ‘AMIP’ AGCMs to ENSO SSTA over 1979-1988; note large differences in midlat teleconnections despite similar tropical Pac response
obs
CCC
SUNYA
MPI
Boyle et al. 1982
3. Regional land surface and cloud feedbacks
Land processes (model-dependent) modify rainfall
Example: Albedo increase from Amazon deforestation
Zeng et al. 1996 Tropical land precipitation is sensitive to albedo (Charney 1975) via a positive ‘convergence feedback’ loop.
Example 2: Regional climate models - NARRCAP
• Each RCM has different physical parameterizations but is driven at boundaries by same global climate model output.
• Loosely interpret as giving precip sensitivity to local physics
50 km regionalclimate models over N America
Global AOGCMs
Courtesy Linda Mearns
Ideally, all RCMs should have same precip trends
• Winter: decent agreement for 2041-2070 minus 1971-2000
Courtesy Linda Mearns
Summer: poor agreement for 2041-2070 minus 1971-2000
Model land and atmosphere representation uncertainties matter more in midlat summer
Courtesy Linda Mearns
Regional cloud trends are circulation-driven
• Clouds, like precipitation, are affected by vertical motion
• Clouds affect the surface and atmospheric energy balance to positively feed back on atmospheric circulations.
• Clouds are challenging for climate models to simulate
• Clouds also dominate uncertainty in global climate sensitivity, which affects the amplitude of precip trends.
More ascent = more deep cloud
Bony et al. 2004
Regional cloud vs. precipitation trendsCloud cover
4. Precipitation uncertainty due to global warming uncertainty
The climate sensitivity problemAll other things being equal, GHG-induced precip trends should scale with global temperature rise, which depends on uncertain emissions and has model uncertainty.
IPCC 2007
For a given scenario, global precip increase scales with model-simulated global warming
Held and Soden 2006
2% K-1
Rainfall trends depend on scenario via global ΔT
• A climate model will warm 4x more and give 4x large precip change for A2 than commitment scenario
CMIP3 multimodel means
Regional precipitation trends hard to model:
1. Tropical precipitation, circulation and SST depend on poorly modeled processes such as cumulus convection.
2. Teleconnection of tropical precipitation changes to midlatitudes also varies between models.
3. Land surface and clouds feed back on precipitation changes4. Precipitation trends modulated by overall global warming, so
dependent on emissions changes and climate sensitivity.
Bottom line: Strong internal feedbacks in the climate system make precipitation trends very sensitive to model details.